Abstract training data.

Abstract training data.

Evolution of cosine-similarity between various words throughout training.

Even a logistic regression quickly gets near 100% accuracy on the validation set when a single blank space is inserted at the end of all viXra text.

Predictions by time step (batch) for a set of randomly selected titles from the validation set. Ordered by distance from .5, i.e. in order of model certainty. arXiv = 0, viXra =1.

viXra = 1, arXiv = 0.

Balanced title data.